🏛️ Citadelle Vision

An Autonomous Multimodal UI Navigator for Legal Research

Citadelle Vision is a next-generation AI agent built for the Gemini Live Agent Challenge. It uses Google Gemini Multimodal to "see" and navigate complex web interfaces (like CourtListener, Oyez, and YouTube), turning 5 hours of manual legal research into a 30-second exportable executive brief.

🚀 Features

• Visual DOM Interaction: Uses Playwright to inject bounding boxes into the UI, allowing Gemini to seamlessly click, type, and extract data using computer vision.
• Autonomous Navigation: Solves complex multi-step tasks (e.g., finding a specific precedent case) without relying on rigid API endpoints.
• Multimodal Extraction: Capable of reading embedded PDFs, transcribing YouTube videos, and scraping complex HTML dynamically.
• Cloud Native: Fully containerized and deployed on Google Cloud Run.

🛠️ Tech Stack

• AI Model: Google Gemini Multimodal (via @google/genai SDK)
• Agent Framework: Playwright (Headless Chromium)
• Backend: Node.js, Express, WebSockets
• Frontend: React, Vite, Tailwind CSS
• Infrastructure: Docker, Google Cloud Run

---

💻 Spin-up Instructions (Local Development)

To run Citadelle Vision locally, follow these steps:

1. Prerequisites

• Node.js (v18 or higher)
• npm or yarn
• A valid Google Gemini API Key

2. Clone the Repository

git clone https://github.com/SifraDev/citadelle-vision-agent.git
cd citadelle-vision-agent

3. Install Dependencies

npm install
# Ensure Playwright browser binaries are installed
npx playwright install chromium

4. Environment Variables

Create a .env file in the root directory and add your Gemini API Key:

GEMINI_API_KEY=your_gemini_api_key_here

5. Start the Application

npm run dev

---

☁️ Google Cloud Deployment

This project is fully containerized and designed to run on Google Cloud Run. The included Dockerfile utilizes the official mcr.microsoft.com/playwright:jammy base image to ensure the headless browser has all necessary system dependencies to execute the agentic loop in a serverless environment.

Deployment Steps:

1. Connect this GitHub repository to Google Cloud Build.
2. Select Dockerfile as the build type.
3. Set the service to use 2 to 4 GiB of Memory and 2 CPUs (required for Playwright).
4. Add the GEMINI_API_KEY to the Cloud Run Environment Variables.
5. Deploy.